As enterprises worldwide race to harness the potential of artificial intelligence, a critical question emerges: How can businesses ensure their AI deployments remain ethical, transparent and trusted? EITN sat down with SAS’s SVP of global marketing, Patrick Xhonneux, to zoom further into SAS-IDC findings and glean exclusive insights.
Trust in AI, trust in algorithms
According to the SAS and IDC report, which surveyed over 2,300 organisations globally, businesses are experiencing what Patrick termed the “trust dilemma.” “Our research showed that while 47% of companies have a very high perception of trust in their AI systems, the actual trustworthiness of the AI systems – when scientifically measured – lags behind,” he revealed. ”
Why is trust in AI at the level that it is?
This psychological trust, he explained, is particularly evident with generative AI: “People naturally trust answers that sound empathetic and confident – even if they know that it is prone to errors.”
He also pointed out that interestingly, people would trust generative AI three times more than machine learning algorithms, because of the conversational nature of the former.
At SAS, Patrick explained, generative AI and predictive analytics are seen as complementary.
“Generative AI or Large Language Models (LLMs) can scan through huge amounts of data, but when you ask generative AI the same question twice, you will get two different answers. This cannot happen for enterprises.”
SAS believes there can be a happy co-existence of two worlds – LLMs that provide ease of access and conversational aptitude that people are comfortable with, while the other, machine learning algorithms, provides predictive, explainable, and transparent solutions.
Patrick cautioned, “Gen AI is wonderful, but you need to understand how it works. It’s a probabilistic model. If you don’t manage to combine both the predictive modelling that gives certainty and explainability of the models, there is a danger.”
Governance – the foundation of ethical AI
The backbone of a trustworthy AI system, Patrick emphasised, is strong governance. And the SAS-IDC report highlighted this. “Enterprises that have a clear governance model in place – those who really understand the need to create trust with the output of their AI – see higher ROI,” he stated.
Typically, enterprises that want to avoid regulatory issues and customer friction, see the importance of this clear governance model, and Patrick called out that ROI from AI can be proven with use cases today.
Gen AI is wonderful, but you need to understand how it works. It’s a probabilistic model. If you don’t manage to combine both the predictive modelling that gives certainty and explainability of the models, there is a danger.
Patrick Xhonneux
How does one kickstart trying to harmonise generative AI applications with machine learning algorithms?
He spoke of the SAS AI governance assessment that helps organisations pinpoint their maturity across culture, operations, and oversight, providing a springboard for scaling AI responsibly. He was also vehement that without a governance model, or a needed safeguard and understanding of an AI investment, enterprises “…will have a nice pilot and it will never scale up.”
The report found that countries like Singapore, Thailand, and Malaysia show high ambition and impact with AI, but still face significant challenges. “These countries have an AI trust index of just 2.93 out of 5 which is pretty low, and a high impact index of 3.3. There’s immense potential for governance to ensure they scale up, and that’s where we think we can help.”
Agentic AI – evolution with oversight
Does agentic AI require rebuilding governance models aka trust from scratch? According to Patrick, agentic AI coordinates multiple LLMs instead of just one, to enable complex workflows and decision making, and sometimes autonomously.
Here the VP explained that SAS has a tool that enables agentic AI to work in an explainable and transparent way. “It’s very visual,” he said, adding that it displays the different steps taken to get to an outcome in an agentic AI workflow.
Making AI explainable
Building explainability into AI deployments isn’t just best practice; it’s essential for auditability and compliance. Two years ago, the advanced analytics company created model cards which are like nutrition labels for AI that look at aspects like fairness, bias, and model performance.
“It will give a visual score of how fair your model is.” he said explaining that this feature is well received internally and externally by customers because of the transparency and auditability that it offers.
A call to action for C-suites
So what should enterprise leaders do now to ensure their AI remains ethical and value-driven? The VP’s advice is pragmatic:
“Start by understanding where you are in your AI creation and deployment. To do that means understanding how to put together that governance model we spoke about.”
In essence, transparency has to be built-in, as well as the best of machine learning algorithms and generative AI combined, while ensuring that there is overall human oversight.
